Showing papers by "Telkom Institute of Technology published in 2013"

A rainfall forecasting using fuzzy system based on genetic algorithm

Abstract: Weather forecasting information is very crucial in decision making process regarding to activities and works, such as in the field of agriculture to determine initial growing season. Recently, climate change causes trouble in weather forecasting. In this paper, rainfall forecasting system using fuzzy system based on genetic algorithm (GA) is made. The data used within this research is taken from Indonesian Agency for Meteorology, Climatology and Geophysics (BMKG) for Kemayoran area, Jakarta. Those data include temperature, air pressure, rainfall, solar radiation, relative humidity, and wind speed. Based on experiment result, it can be concluded that the combination of GA and Fuzzy for Kemayoran weather data can produce prediction model with higher than 90% accuracy with several population size and crossover probability. This condition is highly affected by input data and conducted classification. Obtained model can be used in predicting rainfall class for the next day that divided into 4 class of no-rain or mild rain ( 100 mm).

Comparison between fingerprint and winnowing algorithm to detect plagiarism fraud on Bahasa Indonesia documents

Abstract: Plagiarism detection has been widely discussed in recent years. Various approaches have been proposed such as the text-similarity calculation, structural-approaches, and the fingerprint. In fingerprint-approaches, small parts of document are taken to be matched with other documents. In this paper, fingerprint and Winnowing algorithm is proposed. Those algorithms are used for detecting plagiarism of scientific articles in Bahasa Indonesia. Plagiarism classification is determined from those two documents by a Dice Coefficient at a certain threshold value. The results showed that the best performance of fingerprint algorithm was 92.8% while Winnowing algorithm's best performance was 91.8%. Level-of-relevance to the topic analysis result showed that Winnowing algorithm has got stronger term-correlation of 37.1% compared to the 33.6% fingerprint algorithm.

The analysis and implementation of degree centrality in weighted graph in Social Network Analysis

Abstract: Social Network like Twitter, Facebook, Plurk and Linkedin has million users that constantly evolving and has good information both explicit and implicit information that needs to be explored. Social Network Analysis is the study of social networking with an emphasis on mapping relationships, patterns of interaction between user and content information. One analysis that often be explored is centrality measures that are useful to find the important people in a community by representing in a graph. Kretschmer method is one method of measuring centrality. It is the development of the degree centrality, where the centrality measurement is to find influential people not only involves the number of degree of the user, but also involves the relation weights based on the number of interactions. Therefore, representing the weight graph is very necessary in finding the influential people. Kretschmer method is more suitable for the case of twitter. It is because the most important thing to note is the proximity of the user based on the relationship following/followed and the number of tweet interaction such as mention, retweet and reply. Deeper discussion about the analysis of Kretschmer methods and implementation techniques on twitter is the main discussion of this journal.

Microblogging sentiment analysis with lexical based and machine learning approaches

Abstract: The Digital World encounters rapid development nowadays, especially through the proliferation of social media in Indonesia. Twitter has become one of social media with expanded users within every sectors of society. There are so many part both individual as well as organization/enterprise which utilize twitter as tool for communication, business, customer relation, and other activities. Through the twitter's ever-expanding users with those particular purposes, the precise method to effectively and efficiently analyzing opinion-contained sentences become crucially needed. Therefore this research made for method analyzing through lexical based and model based approaches by machine learning to classify opinion-contained tweets using those 2 methods. The tested machine learning method are Support Vector Machine (SVM), Maximum Entropy (ME), Multinomial Naive Bayes (MNB), and k-Nearest Neighbor (k-NN). Based on the test outcome, lexical based approach highly depended on lexical database which became opinion classification matrix. Whilst machine learning approach can produce better accuracy due to its capability in new training data modeling based on outcome model. However, machine learning model based approach depends on various factors in analyzing sentiment.

Prediction of malaria incidence in Banggai Regency using Evolving Neural Network

Abstract: Malaria is an endemic disease in most of area in Indonesia, especially in rural and remote areas. Banggai, one of regencies in Central Sulawesi province, is a high endemic area of malaria with Annual Parasite Incidence (API) in 2010 reached 7.880/00. The incidence and spreading of malaria were influenced by environmental and weather factors, particularly rainfall and temperature. Therefore this study would like to develop a malaria incidence prediction system based on environmental and weather factors, so that it may assist Indonesian Ministry of Health to control malaria. The method used to solve the problem was Evolving Neural Network (ENN). This method was a mixture between Artificial Neural Network (ANN) and Genetic Algorithm (GA). The result of this study shows that the prediction system has acceptable performance for predicting malaria incidence based on weather factors. The best performance in predicting malaria incidence in 2008 was 21.3% MAPE, 75% accuracy, and 84.21% F-value. While in predicting malaria incidence in 2009 was resulted 15.29% MAPE, 75% accuracy, and 40% F-value. These findings proved that there was a sufficient correlation between weather and malaria incidence. ENN also improved the performance of ANN up to 14.84% in MAPE, 25% in accuracy and 40% in F-value.

Electrocardiagram monitoring on OpenMTC platform

Abstract: Information and Communication Technologies are playing major roles in creating Smart environments, and enhancing services in various sectors. The Machine-to-Machine (M2M) communication paradigm enables devices used in all live aspects to communicate and share information with limited human intervention. In this paper we present an M2M telemedicine service to monitor Electrocardiagram (ECG) signal remotely. ECG is a physiological signal produced by electrical activity of heart; it can be recorded using electrocardiograph device and used to detect many heart problems, such as heart attacks. Data generated by ECG's device are sent over M2M platform, that is OpenMTC for further processing in devices that are on the client side.

Implementation of feature extraction based hand geometry in biometric identification system

Abstract: Biometric identification system is one method that aims to identify a person's identity automatically based on biometric characteristic. Biometric characteristics may include iris, fingerprint, face, voice, or palm. Palm is one of the biometric characteristics that can be used to distinguish a person's identity, because everyone has different palm lines, shapes and sizes. Therefore, in recent years a lot of research done related to biometric identification system based on palm. One of the main problems in the biometric identification system based on the palm is how to extract features automatically without using special devices, because it is the core of biometric identification system. Several information that can be used as the based of this system are the geometry and the lines of the palm. In this study, the palms feature extraction based on it's geometry to find information will be examined, which includes finger width, finger length, palm width, and the ratio between the length of the middle finger, index finger and ring finger. By combining this information can be obtained a characteristic of palms that can be used to recognize a person. The accuracy obtained in this system is 88.4% using the 23 Characteristics of palm geometry without normalization and with 0.035 thresholds on 40 individuals. The characteristic that is used were the fingers length, finger width, and the width of the palm it self. In this experiment, the reduction of characteristics used has a great influent on the accuracy of the system. The test results showed the best accuracy at 35 individuals, with 90.05% accuracy.

Real-time hand-tracking on video image based on palm geometry

Abstract: There are various ways to make the interaction, interaction by using conversation or sign language that has been agreed before. Interaction with sign language especially that use a hand, is tried by people to be implemented to interact with the machine, in this case the machine is computer. In other that, to get the information, the computer must have the ability to track the hand that intended to express the information. This research about hand tracking based on geometry features, by using modified Competitive Hand Valley Detection (CHVD) algorithm to get the reference point. Then the image of the palm of the hand is given a marker on the reference point from mapping result for identifying the shape descriptor based active fingers. From the test results shows that, by using the angle parameter, the application can recognize the shape of the palm of the hand with an average accuracy of 74% in normal conditions, while in extreme conditions obtained an average accuracy of 16%. By using the parameter range, in normal conditions obtained an average accuracy of 2%, while in extreme conditions obtained an average accuracy of 8%. By merging these two parameters, the normal condition the average accuracy obtained 88%, while in extreme conditions obtained an average accuracy of 38%. In terms of speed, the application can run at speeds equivalent to 0.5966 fps 1.6762 seconds or the time it takes to process one frame.

Mobile Tracking System Based on Event Driven Method

Abstract: Monitoring system can be performed both offline and online, and this research carries out the design of long-distance monitoring system for moving system. GPS module is used as system input containing position information of monitored system. This information is processed by the system and is sent by GSM module. On the receiver side, the data are compiled for monitoring system movement. The system is implemented using µC Cortex-M4 by applying Event Driven method. This method is applied for saving resource. The result shows good accuracy level which is measured based on monitoring with system position in the field

Implementation and analysis of remote sensing payload nanosattelite for deforestation monitoring in Indonesian forest

Abstract: Nanosatellite is one type of artificial satellites are relatively small and lightweight mass, between 1 kg - 10 kg. Nanosatellite can be programmed to perform shared mission, one of which is for remote sensing. Remote sensing can be used for many purposes like to learn growth of a region, to create maps, to check weather condition, astronomy monitoring and etc. One of the remote sensing purposes that use in this research is for forest monitoring. Forest area in Indonesia is 120,084,434.79 Ha. To support this mission, need an imaging payload which is equipped with a camera to take the pictures of the object and a microcontroller to process the images before transmitted to the earth station. CAM130 cameras with OV9650 sensor was chosen because it has low prices and small power, so CAM130 is suited with nanosatellite environtment. For microcontroller-based, mini2440 with Samsung S3C2440A processor is selected to process the image. CAM130 and mini2440 this produces an image area coverage of 305.38 km and 240.81 km km based on testing on a map scale. The ability to take pictures and send them relatively quick 2.2 seconds to 3.2 seconds. Remote sensing payload has a total mass of 82.79 grams. CAM130 camera module has dimensions of 3.5 cm × 2.1 cm and mini2440 has dimensions 10 cm × 10 cm. Power consumption of 823 428 mW when the camera is active. The end result of a prototype imaging system for monitoring payload on nanosatelit deforestation in forests in Indonesia in the form of mini-plan map scale 1: 80000 cm.

Development methods for hybrid motion detection (frame difference-automatic threshold)

Abstract: In general the motion detection method look for differences between some sequentially ordered frame followed by threshold process to detect whether occurred any movement or not. Process of deciding threshold to show any movement in dynamic environment has high enough difficulty related with many “fake” movement which appears as the result of illumination change factor, sunshine or light source, color differences as the effect of automatic color correction inside the device, shadow movement, etc. Temporal mechanism using standard deviation is a mechanism that proposed in Video Surveillance and Monitoring (VSAM) research project at Carniagie Mellon Universitiy. In this research we proposed one hybrid scheme to determine initial value of motion thresholding to create an adaptive moving detection system. We proposed the using of local standard deviation in 3 x 3 pixel area as initial value for mechanism that was defined by Robert et al. This hybrid scheme used moving standard deviation value in each pixel to create adaptive threshold value for motion/ movement detection. The experiment showed that by using local standard deviation as initial value we could achieve balance accuracy for movement and non-movement detection with accuracy propostion 80.00% for video with movement and 90.91% for video without movement.

Implementasi dan analisis performansi raid pada data storage infrastructure as a service (iaas) cloud computing..

Abstract: Dalam rangka untuk memaksimalkan utilisasi hardware server, dapat digunakan teknologi cloud computing. Teknologi ini, memungkinkan utilisasi yang tidak terpakai dapat digunakan untuk beberapa virtual server lainnya. Dengan media penyimpanan cloud computing yang sangat besar, kemungkinan terjadinya kesalahan atau kegagalan menjadi tinggi, karena banyaknya data yang akan disimpan atau diproses. Akibatnya, kinerja (performansi) disk pun menjadi menurun. Salah satu cara untuk meningkatkan kinerja (performansi) disk adalah dengan menggunakan RAID ( Redundant Array of Independent Disk ). Dengan diterapkannya RAID pada data storage Infrastructure as a Service (IaaS) Cloud Computing, dari sisi performansi, RAID 0 dapat meningkatkan kecepatan baca sebesar 11.13% dan kecepatan tulis sebesar 23.35%, dibandingkan dengan tanpa RAID. Sedangkan untuk nilai IOPS ( Input-Output Operation Per Second ) terjadi peningkatan sebesar 74.80% pada saat sebelum resource digunakan oleh instance dan 80.89% pada saat setelah resource digunakan oleh instance , dibandingkan dengan tanpa RAID.

Fire color detection using color look up and histogram analysis

Abstract: Detecting the presence of fire in an image or video is one of the domain problem in computer vision. Visual based fire detection occurs to overcome the weakness of the conventional, specially in monitoring area. There has been a lot of proposed method to detect the presence of fire in image or video [2,3,4,5,6,9]. Based on proposed method, the process to detect presence of fire in image or video can be considered as multi-filtering process. One of the phase for fire detection is to determine candidate fire based on their color. In this research we focuses on determining the presence of fire in image based on their pixel intensity and their distribution. Fire has unique color that can be used to distinguish them from other objects. There are several ways to determine candidate fire pixel based on their intensity property like using chromatic rule [3] and color look up table [14]. We proposed additional process using simple approach based on value of pixel distribution (histogram analysis). Our experiment shows that by using color look up table combined with histogram analysis, we can determine the presence of fire in image with accuracy about 100% for fire image categories and 60% for fake fire image categories.

Design and realization of two array triangle patch of microstrip antenna with gold plat at frequency 2400–2450 MHz for hexagonal nanosatellite

Abstract: Ministry of Education and Culture developed a nano-sized satellite technology, or so-called nanosatellite, to design Indonesia inter-University Satellite-1 (IiNUSAT-1) as a learning tool of space engineering, for the universities in Indonesia. These nano satellites orbiting in Low Earth Orbit trajectory (LEO). This satellite has a primary function for data communication. On the space segment subsystems are RSPL (Remote Sensing Payload) as an image sensor payload following the transmitter system (antenna) that can be used for sensing applications earth's surface. Based on the above conditions, this research was made of two arrays of microstrip antenna with a triangular patch. This antenna works on S-band frequency, on 2400-2450 MHz with VSWR ≤ 1.7. To meet a good data transmit ability, the antenna has been designed with gain above 6 dBi, with the transmitting range of 700 km. On the surface of the patch antenna, will be given a gold plating to extend the life of the antenna from corrosion. From the measurement results, obtained values of VSWR is 1.150 at center frequency in 2.425 GHz for the antenna with the gold plat. Gain obtained from the measurements is 6.120 dBi. This antenna has a transmit pattern of unidirectional and polarized elliptical, circular approach.

Mining food industry's multidimensional data to produce association rules using apriori algorithm as a basis of business strategy

Abstract: The food industry sell a range of product variations. The company want to take advantage to build business strategy from huge information which is stored in data warehouse. In this case, data mining technology needs to be implemented to explore valuable information on transactional data to assess customer's preferences for products sold as a business strategy.

Analisa Sinyal EKG dengan Metoda HRV (Heart Rate Variability) pada Domain Waktu Aktivitas Berdiri dan Terlentang

Abstract: Tingginya angka kematian di Indonesia akibat penyakit jantung koroner (PJK) mencapai 26%. Tingginya angka tersebut, mengakibatkan PJK sebagai penyebab kematian nomor satu di Indonesia. Oleh karena itu, untuk mencegah tingginya angka kematian PJK, dapat dilakukan pemeriksaan rekaman EKG secara dini. Hasil rekaman EKG dianalisa dengan menggunakan metoda HRV pada domain waktu untuk mengetahui aritmia jantung seseorang sehingga dapat mengambil langkah medis yang tepat. Analisa EKG yang diteliti saat aktivitas berdiri dan terlentang pada 5 subyek yang terdiri dari 4 laki- laki dan 1 perempuan yang berumur antara 20 tahun sampai 58 tahun dalam waktu 5 menit. 1. Hasil analisa metoda HRV pada domain waktu dengan membandingkan aktivitas berdiri dan aktivitas terlentang: kecepatan rata- rata detak jantung semakin menurun, waktu rata- rata interval RR semakin meningkat, waktu minimum interval RR semakin meningkat, waktu maksimum interval RR semakin meningkat, standar deviasi, nn50, dan pnn50 meningkat. Hal ini menunjukkan perubahan aktivitas dari berdiri ke terlentang dapat memperlihatkan perubahan aktivitas simpatik ke parasimpatik pada metoda HRV pada domain waktu.

Identifikasi Telapak Tangan Menggunakan 2DPCA plus PCA dan K-Nearest Neighbour

Abstract: Biometrik merupakan pengenalan seseorang berdasarkan anatomi ataupun karakteristik tingkah laku. Identifikasi telapak tangan sebagai teknik biometrik memiliki kelebihan yaitu membutuhkan resolusi citra yang kecil, biaya alat yang murah, tidak bersifat intrusif, serta memiliki tingkat akurasi yang tinggi. Paper ini akan membahas identifikasi telapak tangan menggunakan metode 2DPCA plus PCA. Principal Component Analysis (PCA) merupakan teknik ekstraksi fitur yang memiliki kelebihan dari segi kompresi data. Metode 2 Dimensional Principal Component Analysis (2DPCA) memiliki kelebihan dari PCA dari segi akurasi dan kompleksitas waktu, namun memiliki kekurangan yaitu membutuhkan banyak koefisien dalam menyimpan data. Penggabungan kedua metode tersebut menghasilkan metode yaitu 2DPCA plus PCA. Pada percobaan, klasifikasi dilakukan dengan menggunakan K-Nearest Neighbour (KNN). Analisis dilakukan dengan membandingkan performansi metode 2DPCA plus PCA dengan metode lainnya yaitu 2DPCA dan PCA.

ACTIFIST: Adaptive architecture for integrated information system

Abstract: The need of Integrated Enterprise Information System in nowadays business is obvious. Not only to support daily operation, Integrated Enterprise Information System is also crucial in management and providing valid information to strategic decision. Business often changes based on circumstances, it drives enterprise to do some adjustments in its internal system to keep competitive. On this situation, changes in Integrated Enterprise Information System are unavoidable. This is the main reason why enterprise should adopt architecture that enables its Integrated Enterprise Information System to be adaptive and customable, while still assures integration.

Implementation of networked control systems using programmable controller based Ethernet network

Abstract: This paper presents the implementation of networked control systems (NCS) to be applied in industrial automation by using the programmable controller as the platform controller based Ethernet network. The fast growing in ICT technology on recent era brings some essential progress to support the development of NCS becomes more reliable to be implemented especially for critical delay application in industrial automation. The experiment, which is to measure the delay on the control signal that occur, the control execution accuracy and the cycle time of the programmable controller, had been conducted in four phase from identifying component of NCS and its characteristics, designing NCS, taking simulation of process workflow and simulation of the network and implement to physical system. The delay measure by monitoring the status of communication port of programmable controller. The measurement of control execution accuracy took after observation of process of removing the box from conveyor. The programmable controller's cycle time measurement obtained by capturing the cycle time data from its memory. The experiment results shows that the latest programmable controller along with recent network protocols has capability to deal with the delay and traffic. Delay in the network could decrease the control execution accuracy. The best control accuracy performances occur when total delay in the network are less than 1 s. Thus, the critical delay <; 1 s. In other hand, the traffic condition does not have any impact related to the cycle time.

An analysis comparison of AODV UU and batmand performance for mobile ad-hoc network

Abstract: Nowadays, technology tends to be developed to wireless technology. One of factors affects that development is mobile capability offered by the system to users. One of current wireless technology is mobile ad-hoc network. Important feature of mobile ad-hoc network implementation is routing protocol such as AODV and BATMAN. Implementing AODV and BATMAN in real life is by using AODV UU and BATMAN Daemon. In this paper, four nodes are implemented to evaluate both routing protocols in mobile ad-hoc network. For mobile scenario, source node moves with velocity about three km/h back and forth to destination node. Sending and receiving files in system is using Distributed Internet Traffic Generator (DITG). Analyzed parameters in this system are average packets loss percentages, average throughput, and average jitter. Implementation of no movement scenario shows that average packets loss percentage ranges from 0% to 3.358%, average throughput ranges from 393.1821813 Kbps to 409.9955152 Kbps, and average jitter ranges from 0.00052182 s to 0.00604264 s. In addition, implementation of source node movement scenario shows that average packets loss percentage ranges from 0.002% to 4.276%, average throughput ranges from 386.6820413 Kbps to 409.8114591 Kbps, average jitter ranges from 0.00202616 s to 0.00609198 s.

Techno-economic and regulation impact analysis of Mobile Number Portability implementation

Abstract: Mobile Number Portability (MNP) is an ability of changing operator (porting) without changing the number. This scheme provides freedom and convenience for customers to get the best services. This scheme allows the regulator to gain efficiency of block number as a limited resource. Analysis in this research consisted of techno-economic analysis to provide options for the operator (Telkomsel) and RIA (Regulatory Impact Analysis) for the regulator. The output from these two analyses were processed and simulated based on customer behavioral data to obtain the impact of MNP. Both analyses indicated that market operators did not change drastically after MNP implementation. On Telkomsel, number of customers who port-in exceeds the port-out (48% against 43%). RIA analysis showed that the benefits received by the public exceeded the cost incurred.

Copyright Protection in Audio File Using Watermarking Approach Based on Wavelet-SVD

Abstract: The encryption is not be able to protect intellectual properties of audio file completely. Therefore, a technique of audio watermarking is launched to improve the previous technique of the protection. In this paper, we analysis and implements a system that be able to protect a copyright of an audio file by using wavelet-SVD based on watermarking Approach. IWT-SVD produces the best rate of inaudibility compared with DCT-SVD and DWT-SVD, but total processing time of watermarking using DWT-SVD is faster than others.

The optimization of maintenance time and total site crew for Base Transceiver Station (BTS) maintenance using Reliability Centered Maintenance (RCM) and Life Cycle Cost (LCC)

Abstract: In a telecommunications network architecture, BTS holds a very important role. In the event that causes disruption of BTS and BTS to be down, it will result in loss of potential revenue and lead to loss of consumer loyalty in the company. Hence the need for an effective method of treatment BTS and BTS efficient so that availability can be maintained properly. The method used in the treatment of BTS in this study is the RCM method (Reliability Centered Maintenance) to determine treatment priorities of existing equipment in the BTS based on vital functions and data destruction in order to obtain its value MTTF and MTTR, as well as LCC method (Life Cycle Cost) to determine the optimal number of sites crew for treatment BTS. By using the RCM method, the results of the determination of the critical components based on damage data, are obtained two set of equipment as critical components. They are generator and transmission module BCKM. Calculation and data processing of field data acquisition and secondary data the previous year, the value of the MTTF BCKM module is 1109.635 hours and MTTR BCKM 6,095 hours, while the value of the generator is 1647,373 hours MTTF and MTTR generator is 3.6 hours. While the number of site crew required in the BTS treatment based on the results of calculation and data processing with the LCC method, obtained the most optimal amount based on the cost is 4 site crews.

Design of integrated scheduling and automated controlling for surface treatment process using Supervisory Control and Data Acquisition (SCADA)

Abstract: This paper aims to design a model of integrated scheduling and automated controlling for a surface treatment process in an aircraft manufacturing company, where in this case aircraft parts are dyed into chemical liquid to be corrosion-proof. In the existing plant, parts dyeing processes (moving the crane and monitoring the tanks temperature) are done manually. Integrated scheduling and automation control system is proposed to improve planning, control and safety issues in the dyeing process. The scheduling system is developed using fuzzy logic algorithm to minimize makespan and the automation system is designed based on Supervisory Control and Data Acquisition (SCADA). A case study has shown that the simulation of proposed model can result in reducing makespan of about 51,3 % and the schedule can be implemented automatically in the surface treatment prototype.

Combining pixel projection, OCR, and scale calculation to perform graph feature extraction

Abstract: Graph Feature Extraction is the task to collect any important feature on the graph image so the extracted data can be further processed. In this paper, it is assumed that the graph consists only one single curve. The extracted features here are graph title, axis title, axis values, legend, scales, and curve values.

Multi agents based traditional market customers' behavior design

Abstract: Mismanagement in traditional market has made its existence was harmed by modern market. Traditional market is identified as a dirty, crowded, and uncomfortable place to shop. This condition is very different with the condition in modern market. One of the challenges is making the traditional market become a comfort place to shop. The research about crowd modeling in traditional market is being done. One of the partial researches in this main research is designing traditional market customers' behavior. Multi agents system has been chosen as a platform because it has been chosen in many crowd modeling researches. This paper is a report of this research. This research process is divided into steps. The first step is reviewing literatures which have correlation with the research topic. The second step is doing field observation. The field observation has been done in one of traditional market in Yogyakarta. The third step is designing the customers' behavior based on the literatures and field observation. The fourth step is developing simulation based on the design. From the simulation, it can be found that there is potential where persons can't pass through the corridor because the corridor is fully loaded. Some solutions are proposed to avoid that situation.

Analisis Prediksi Churn Dan Segmentasi Pelanggan Speedy Retail Daerah Operasional Bandung Menggunakan Algoritma Decision Tree Dan K-Means

Abstract: Kecenderungan penurunan jumlah pelanggan Speedy pada PT. Telekomunikasi Indonesia Tbk. (Telkom), khususnya pelanggan Speedy retail daerah operasional Bandung perlu diantisipasi. Untuk dapat bersaing dengan berbagai ISP (Internet Service Provider) yang menawarkan produk dengan keunggulan tertentu, maka Telkom membutuhkan program retensi pelanggan agar pelanggan tidak memutuskan layanan (churn) dan beralih ke provider lain. Penelitian ini dilakukan dengan menerapkan data mining. Terdapat dua pemodelan dalam penelitian ini, yaitu prediksi churn menggunakan algoritma Decision Tree, serta segmentasi pelanggan menggunakan algoritma Clustering K-Means. Pemodelan prediksi churn menggunakan algoritma Decision Tree menghasilkan pelanggan yang berpotensi churn dengan variabel-variabel terpilih yaitu Paket Layanan, Customer Billing, Usage, Durasi Berlangganan, Jumlah Gangguan, dan Status Churn. Berdasarkan pola pemodelan prediksi churn, akurasi yang dihasilkan adalah 99,47%. Kemudian dilakukan segmentasi pada pelanggan yang berpotensi churn menggunakan algoritma Clustering K-Means, untuk menemukan sasaran pelanggan yang akan diberikan program retensi. Program retensi yang akan ditawarkan dalam penelitian ini berupa cross selling dan penawaran layanan pelengkap. Untuk layanan cross selling, akan ditawarkan Groovia TV kepada pelanggan sasaran program retensi, yang telah memenuhi persyaratan jaringan untuk pemasangan layanan. Setelah dilakukan analisis, dari 1912 sasaran pelanggan hasil segmentasi, terdapat 601 pelanggan yang memenuhi persyaratan penawaran Groovia TV dan layanan pelengkap, serta 1311 sasaran pelanggan layanan pelengkap saja.